Update to Unabyss

{
  "name": "brand-voice",
  "description": "Derive a founder or company voice profile from real writing samples found in connected sources, then use it to ensure all content sounds like the actual person — not generic AI copy. Use this skill whenever the user asks about their writing style, brand voice, tone of voice, how they sound, or wants any content to match their voice. Also trigger when another skill needs a voice profile before writing copy, posts, emails, or outreach.",
  "included_files": [],
  "skill_md_contents": "---\nname: brand-voice\ndescription: \"Derive a founder or company voice profile from real writing samples found in connected sources, then use it to ensure all content sounds like the actual person — not generic AI copy. Use this skill whenever the user asks about their writing style, brand voice, tone of voice, how they sound, or wants any content to match their voice. Also trigger when another skill needs a voice profile before writing copy, posts, emails, or outreach.\"\n---\n\n# Brand Voice\n\nExtract a reusable voice profile from the founder's actual writing — so every piece of content sounds like them, not like a language model approximating them.\n\n## Required Integrations\n\nThis skill uses **Unabyss MCP** — `agentic_query` to surface real writing samples across all connected sources.\n\n> If Unabyss is not connected, tell the user: \"This skill requires Unabyss MCP. You can connect it from the Tools menu.\"\n\n---\n\n## How to Run\n\n### Step 1 — Check for an existing export\n\nCall `export_list` first. If a brand voice or voice profile export already exists and is fresh, call `export_read` — no need to regenerate.\n\n### Step 2 — Pull writing samples with `agentic_query`\n\nSamples are scattered across emails, posts, docs, Slack — use `agentic_query`:\n\n```\nFind examples of my own writing from across all connected sources — emails I've sent, LinkedIn or X posts, Slack messages where I'm explaining something, docs or notes I've written, launch announcements, product updates, or any other text where I'm clearly the author. Prioritize recent material. Return the actual text, not summaries.\n```\n\nCall `agentic_query_read` to retrieve results. Aim for 8–12 distinct samples across different contexts.\n\n### Step 3 — Analyze and produce the profile\n\n---\n\n**VOICE PROFILE — [Name / Company]**\n*Generated from [N] samples*\n\n**In one sentence**\nHow this person writes, in plain language.\n\n**Tone**\n3–5 specific descriptors, each with a concrete example from the actual samples.\n\n**Sentence style**\nTypical length, structure, how they open and close ideas.\n\n**How they make claims**\nDirect / hedged / evidence-first / assertion-first — with examples.\n\n**What they reach for**\nNumbers, analogies, product examples, customer stories.\n\n**Vocabulary**\nWords and phrases that appear naturally. Words and constructions they never use.\n\n**What to avoid**\nSpecific patterns, phrases, or tones that would immediately sound wrong.\n\n**3 on-voice examples**\nPulled verbatim from samples, with a one-line note on why each works.\n\n---\n\n### Step 4 — Save the profile\n\nOnce generated, call `export_create_from_text` to save the voice profile as an export — so other skills can load it instantly next time with `export_read` instead of regenerating.\n\n---\n\n## Output Rules\n\n- Always check `export_list` before running `agentic_query` — voice profiles change slowly\n- Every observation grounded in an actual sample — no invented characterizations\n- If samples are thin, note which sources seem missing and produce a partial profile with caveats\n- No preamble — start directly with the voice profile\n"
}